From raw orders to useful analysis.

Two compact demonstrations: a data pipeline that rejects bad input, and an OLAP model that turns clean transactions into analytical questions.

CSV ingestion, validation and Parquet curation, DuckDB OLAP, and GitHub Pages deployment

One dataset, two engineering questions

Can this data be trusted before it reaches analytics? Then, can we summarize it by time, region, category, and product without changing the source records? The examples use synthetic sales rows only.

Data Engineering

Python and DuckDB load a CSV, validate required fields and unique order IDs, normalize the data, and publish Parquet.

Read the pipeline report →

Online Analytical Processing

DuckDB builds a small star schema and computes totals that support slice, dice, and drill-down analysis.

Open the OLAP report →

How this demo is published

A GitHub Actions workflow runs tests, rebuilds both reports, and deploys the static site to GitHub Pages on pushes to main. It also runs weekly. The queries are executed during the build; the website serves static results, not a live database.

View the workflow runs →